There are really two ways to work with large language models, you either rent a closed one through an API or you learn to run and adapt open ones yourself, and this course is the best free path I know into that second world. It used to be called the NLP course and Hugging Face has since rebuilt and renamed it as the LLM course, which is a fair reflection of how much it now covers. The early chapters, one through four, are the foundation, you learn how transformer models actually work and then immediately get hands on with the Transformers library, pulling a model off the Hub, running inference, and fine tuning it on your own dataset with the Trainer API. The middle chapters take you through Datasets and Tokenizers and into the classic language tasks, which is where you stop treating the library as magic and start understanding what it is doing under the hood.
Then come the newer additions that justify the rename, dedicated chapters on fine tuning large language models, on building reasoning models in the style of the recent open releases, and material that reaches into agents, so it is no longer just an NLP primer but a genuine route into modern LLM engineering. Everything is hands on, the notebooks run in Colab so you do not need a powerful machine to start, and it is completely free with no ads, which for the quality on offer is remarkable. The trade off is that this is unapologetically technical, you are reading and writing real Python from the first chapter, so it is not the place to send someone who just wants to use ChatGPT better. The breadth is also a double edged sword, the course is large now and the later chapters move quickly, so if you rush them they can feel dense and you will get more from slowing down and doing the exercises.
And yes, it is built around the Hugging Face ecosystem, but since that ecosystem is effectively the standard for open models, that is a feature rather than a limitation. For anyone serious about fine tuning and deploying their own models, this is where I would start.